Add low accuracy sin, cos, log2, exp2 and pow (#229)

This patch adds the following functions.

* fastsinf_u3500
* fastcosf_u3500
* fastpowf_u3500
* log2_u35, log2f_u35
* exp2_u35, exp2f_u35

The error bound of fastsin and fastcos are max(2e-6, 350ULPs).
Each function has corresponding deterministic functions.